The Exchange Rate and its Fundamentals in a Complex World Paul De Grauwe and Marianna Grimaldi University of Leuven/CME April 2002
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
The Exchange Rate and its Fundamentals
in a Complex World
Paul De Grauwe and Marianna Grimaldi
University of Leuven/CME
April 20021. Introduction
Since the start of the floating exchange rate regime in the early seventies, an increasing
amount of empirical anomalies and puzzles have been uncovered that seem to contradict the
existing exchange rate theories. 1
The first and foremost empirical puzzle has been called the “disconnect” puzzle (see Obstfeld
and Rogoff(2000)), i.e. the exchange rate appears to be disconnected from its underlying
fundamentals most of the time. It was first analysed by John Williamson who called it the
‘misalignment problem’. This puzzle was also implicit in the celebrated Meese and Rogoff
studies of the early 1980s documenting that there is no stable relationship between exchange
rate movements and the news in the fundamental variables. Goodhart (1989) and Goodhart
and Figlioli (1991) found that most of the changes in the exchange rates occur when there is
no observable news in the fundamental economic variables. This finding contradicted the
theoretical models, which imply that the exchange rate can only move when there is news in
the fundamentals.
Other empirical anomalies have been uncovered over the years. One is the puzzle of “excess
volatility” of the exchange rate, i.e. the volatility of the exchange rate by far exceeds the
volatility of the underlying economic variables. Baxter and Stockman (1989) and Flood and
Rose (1995) found that while the movements from fixed to flexible exchange rates led to a
dramatic increase in the volatility of the exchange rate no such increase could be detected in
the volatility of the underlying economic variables. This contradicted the ‘news’ models that
predicted that the volatility of the exchange rate can only increase when the variability of the
underlying fundamental variables increases.
A third puzzle relates to PPP and is closely related to the “disconnect puzzle”. Many
researches have found that the deviations from PPP are large and sustained (Rogoff (1995),
Obstfeld and Rogoff (2000), Cheung and Lai (2000)). The half-life of the PPP deviations has
been estimated to be of the order of 4 to 5 years. Some researchers have found even longer
half-lives (Lothian & Taylor (1998), Engel(2000), O’ Connell (1998)). Other researchers
(Dumas (1992)) have stressed that the long time needed to adjust to PPP might be due to the
existence of transaction costs. The transaction cost hypothesis implies a non-linearity in the
adjustment process. This hypothesis has been confirmed by the empirical evidence based on
time series analysis (Michael, Nobay, Peel (1997), Kilian & Taylor (2001)).
A fourth puzzle is that the distribution of the exchange rate returns does not follow a random
walk. Most of the empirical findings document that the exchange rate returns have fat tails
(see de Vries(2001), Lux T. (1997, 1998), Lux and Marchesi (1999, 2000). This evidence is
difficult to rationalise in existing exchange rate models, since there is little evidence of fat
tails in the fundamental variables that drive the exchange rate in these models.
It is obvious that in the face of these empirical anomalies the existing exchange rate models
that have been developed mostly in the 1970s need to be re-evaluated. The empirical failure
of these exchange rate models has led to new attempts to model the exchange rate. These
attempts have led to three different modelling approaches. The first one uses the Obstfeld–
Rogoff(1996) framework of dynamic utility optimisation of a representative agent. The
models that came out from this approach have a high content of intellectual excitement.
However, they have not yet led to many testable propositions.
1
We are grateful to Yin-Wong Cheung, Ronald MacDonald, Assaf Razin and Marc Flandreau for many interesting comments
and suggestions. We also profited from comments made during seminars at the Nederlandsche Bank, the European Central Bank,
and CESifo Munich.
2A second approach starts from the analysis of the microstructure of the foreign exchange
market with heterogeneous agents (see Evans and Lyons (1999), Lyons (2001)). This
approach has led to new insights into the way information is aggregated and is important for
the understanding of the very short-term behaviour of the exchange rate. As a result, this
approach is capable of explaining some of the empirical anomalies, e.g. the excess volatility.
However, it is as yet unclear whether it can solve the other empirical anomalies discussed
here.
In this paper we use the insights of the microstructure literature recognising that
heterogeneous agents have different beliefs about the behaviour of the exchange rate. The
difference in the beliefs of agents introduces non-linear features in the dynamics of the
exchange rate.2 We develop a simple model of the exchange rate, which incorporates these
non-linear features and we analyse their implications for the dynamics of the exchange rate. It
will be shown that our simple non-linear model is capable of explaining the empirical puzzles
about the exchange rate behaviour.
2. A Simple non-linear exchange rate model
In this section we briefly describe the model. In appendix we present the model in more
formal detail.
The essential features of the model are the following. There are two kinds traders,
fundamentalists and chartists. The fundamentalists compare the market exchange rates with
the fundamental rate and they forecast the future market rate to move towards the
fundamental rate. In this sense they follow a negative feedback rule 3 . We make the
assumption that they expect the speed with which the market rate returns to the fundamental
rate to be determined by the speed of adjustment in the goods market.
As pointed out earlier, there is an increasing amount of empirical evidence indicating that the
speed of adjustment in the goods market follows a non-linear dynamics, i.e. the speed with
which prices adjust towards equilibrium depends positively on the size of the deviation from
equilibrium. We assume that this adjustment process is quadratic in nature4 . Fundamentalists
take this non-linear dynamic adjustment into account in making their forecast.
Thus when the size of the deviation from equilibrium is large the fundamentalists expect a
faster speed of adjustment towards the fundamental rate than when the size of the deviation is
small. The economics behind this is that in order to profit from arbitrage opportunities in the
goods market, some fixed investment must be made, e.g. trucks must be bought, planes be
chartered, etc. These investments become profitable with sufficiently large deviations from
the fundamental exchange rate.
The chartists are assumed to follow a positive feedback rule, i.e. they extrapolate past
movements of the exchange rate into the future. The parameter of extrapolation will be seen
to play an important role in the dynamics of the exchange rate.
Our choice to introduce chartists’ rules of forecasting is based on empirical evidence. The
evidence that Chartism is used widely to make forecasts is overwhelming (see Cheung and
Chinn (1989), Taylor and Allen (1992)). Therefore, we give a prominent role to chartists in
2
It should be noted that the heterogeneity of agents’ expectations has been recognised as being important to explain the dynamics
of asset prices, including the exchange rate (see De Long, et al. (1990), Hommes and Brock (1998), Lux and Marchesi (2000),
Frankel and Froot (1986)).
3
Note that this is also the approach taken in the Dornbusch model.
4
See Kilian and Taylor (2001). See also De Grauwe and Grimaldi (2001) in which we showed that a quadratic specification fits
the data rather well.
3our model. It remains important, however, to check if the model is internally consistent. In
particular, the chartists’ forecasting rule must be shown to be profitable within the confines of
the model. If these rules turn out to be unprofitable, they will not continue to be used. They
can then also not been maintained in the model. We will therefore check whether these rules
are indeed profitable.
A final feature of the model is the existence of transactions costs. We assume that transactions
costs prevent goods arbitrage to function as long as the exchange rate remains within the
transactions cost band. Fundamentalists take the existence of this transactions cost band when
forecasting the exchange rate, i.e. when the exchange rate remains within the band they know
that arbitrage does not operate. As a result, they do not expect to return to its equilibrium
value.
3. The model with random shocks (news)
In this section we investigate the solution of the model when random shocks in the
equilibrium exchange rate occur. The question we analyse in this section is how well our
model mimics the empirical anomalies and puzzles identified in the introduction. We start
with the ‘disconnect puzzle’ and we analyse how the market exchange rate behaves relative to
the fundamental exchange rate in our model.
3.1 The disconnect puzzle
In figure 1 we show the two variables, for a particular combination of parameters. We show
the results of a sensitivity analysis later.
We observe that the market rate can deviate from the fundamental value substantially and in a
persistent way. Moreover, it appears that the exchange rate movements are often disconnected
from the movements of the underlying fundamental. In fact, they often move in opposite
directions. Thus, the model is able to generate an empirical regularity (the ‘disconnect’
puzzle) that has also been observed in reality5 . We will return to this result later to analyse
how sensitive it is to particular parameter values, like speed of adjustment and transaction
costs.
Figure 1
exchange rate and fundamental
c=5 , teta=-0.3 , beta=1.2
7
6
5
fundamental rate
4
3
2
1
0
-1
market rate
-2
-3
1
41
81
121
161
201
241
281
321
361
401
441
481
521
561
601
641
681
721
761
801
841
881
921
961
5
See Obsteld and Rogoff (2000). See also De Grauwe (2000) for a survey of the empirical evidence. In De Grauwe and
Vansteenkiste (2001) we present additional empirical evidence.
4Closely related to the ‘disconnect’ puzzle is the empirical evidence provided by Meese and
Rogoff (1983) in their celebrated empirical tests of exchange rate models. The Meese and
Rogoff analysis implies that the existing exchange rate models do not forecast the frequent
structural breaks in the link between the exchange rate and the fundamentals. We show that
our model is capable of reproducing the Meese and Rogoff result. We implemented this idea
by introducing a small change in the speed of adjustment in the goods market, θ i.e., this
parameter was changed from –0.2 to –0.21. All the other parameters were kept unchanged.
We then simulated the exchange rate with the small error in θ, and compared this with the
exchange rate obtained with the ‘true’ θ (-0.2). We show the result in Figure 2. These
simulations can be interpreted as follows. The simulation using θ = –0.2 generates the
exchange rate produced by the true model. The second simulation can be considered to be the
exchange rate, which is produced by a researcher who has estimated the model and made a
small measurement error. This exchange rate can also be interpreted as the exchange rate
forecast using the estimated model containing a small measurement error. We observe that
the small error leads after some time to a very different time-path of the exchange rate,
producing the appearance of large structural breaks. This result is related to the ‘sensitivity to
initial conditions’ of non-linear models. It should be noted, however, that we obtain this
result, even though we have a parameter combination that does not produce chaos in the
deterministic part of the model.
In the next step we computed the one-period-ahead forecast error made by this researcher who
uses her almost perfect estimate of the true model, and we compared this forecast error with
the forecast error using the random walk. As measures of forecasts errors, we computed both
the mean root squared error and the mean of the absolute deviations. The latter is a better
measure of the forecast error when the exchange rate distribution is not gaussian but shows fat
tails 6 .
We find that the random walk forecast outperforms the forecast made by a researcher who
made a small measurement error estimating the “true” model7 . We present some examples in
table 1. The remarkable aspect of these results is that such a small measurement error is
capable of producing large structural breaks and poor forecasting performance. It is therefore
no surprise that when real-life econometrics is used to estimate models, these models perform
so poorly in forecasting exercises when compared to the random walk forecast.
6
This approach was also followed by Meese and Rogoff (1983).
7
Note that, as in Meese and Rogoff, the researcher using a model to make forecasts has more information available than the
random walk forecaster. The former has perfect knowledge of the one-period-ahead fundamental while the latter does not have
that information.
5Figure 2
Exchange rate with and without small change in theta
c=5; theta=-0.2;beta=1.5
15
10
5
0
-5
-10
theta=-0.21
theta=-0.2
-15
1
25
49
73
97
121
145
169
193
217
241
265
289
313
337
361
385
409
433
457
481
Table 1: Measures of forecast errors using an ‘econometric’ model and the random
walk: 1-period ahead forecasts
Root mean squared errors Mean absolute deviation
model random walk model random walk
Sample 1-1000 3.17 0.78 1.57 0.75
Sample 1001-2000 2.31 0.81 1.29 0.77
Sample 2001-3000 2.78 0.80 1.50 0.76
Another interesting aspect of these results is that the poor performance of model-based
forecasts is not due to the fact that there are regime switches. Economists have been tempted
to interpret the Meese and Rogoff results as reflecting changes in regime that lead to
parameter instability. The problem with that explanation is that one has to assume too many
regime switches to explain the parameter instability. There are just not enough regime
switches around. Our model shows that one can have the appearance of frequent structural
changes without regime switches. These structural changes arise from the non-linearities in
the model.
We extended the analysis for n-period ahead forecasts. Meese and Rogoff found that when
forecasts are made over longer periods, the forecasts based on econometric models become
more useful. We find the same result in our model. We computed the 5-period and 10-period
ahead forecasts using the ‘econometric model’ and using the random walk. The results are
presented in table 2. We observe that when the forecasting horizon increases to 10 periods the
model-forecasting becomes better than the random walk. It should be stressed, however, that
as in Meese and Rogoff we load the dice in favour of the model forecaster by assuming that
he has perfect knowledge of the fundamental variable ten-periods in the future. No such
strong informational advantage is assumed when using the random walk forecast.
6Table 2: Measures of forecast errors using an ‘econometric’ model and the
random walk: 5-periods and 10-periods ahead forecasts
Root mean squared errors Mean absolute deviation
model random walk model random walk
5-period ahead forecasts
Sample 1-1000 3.17 2.02 1.57 1.28
Sample 1001-2000 2.31 2.18 1.29 1.31
Sample 2001-3000 2.78 2.10 1.50 1.29
10-period ahead forecasts
Sample 1-1000 3.18 3.48 1.57 1.68
Sample 1001-2000 2.31 3.66 1.29 1.71
Sample 2001-3000 2.78 3.63 1.50 1.71
3.2. The disconnect puzzle: sensitivity analysis
In this section we investigate the sensitivity of the results relating to the ‘disconnect puzzle’ to
changes in the value of the transaction costs band. In particular, in the next figure 3 we
compare the movements of the market exchange rate under two assumptions about
transaction costs. In panel a we assume that transaction costs are zero and in panel b we
assume them to be high, i.e. equal to 5.
Figure 3a Figure 3b
market rate anf fundamental rate with no transaction costs market rate anf fundamental rate with high transaction costs
(c=5 , teta=-0.2 , beta=1.5)
1 6
5
0
fundamental rate
market rate
4
-1
3
-2
fundamental rate
2
-3
1
market rate
-4
0
-5 -1
-6 -2
1
20
39
58
77
96
1
20
39
58
77
96
115
134
153
172
191
210
229
248
267
286
305
324
343
362
381
400
419
438
457
476
115
134
153
172
191
210
229
248
267
286
305
324
343
362
381
400
419
438
457
476
The contrast between the two panels is striking. When transaction costs are zero, the market
exchange rate does not deviate substantially for a long period of time from its fundamental
7value. In contrast, when transaction costs are present the deviations of the market exchange
rate from its equilibrium value are large and persistent. Thus, transactions costs in the goods
markets are important in explaining the disconnect puzzle (see Obstfeld&Rogoff (2001)).
Transaction costs have also other important implications for the dynamics of the exchange
rate. We show this in figure 6, where we introduce a negative and permanent shock (-0.01) in
the fundamental exchange rate. Thus, over time the “new” fundamental exchange rate
progressively but slowly departs from the “old” one.
Figure 4 allows us to see how this accumulating small change in the equilibrium value of the
exchange rate, which occurs in period 1, leads to a large jump in the exchange rate many
periods later. This change has the appearance of a regime shift in spite of the fact that the
change in the fundamental exchange rate is very small and continuous. This feature is much
related to the existence of transaction costs, which implies that the effect of the accumulated
changes in the fundamental exchange rate will be visible only when it will overcome the
transaction costs band.
Figure 4
sensitivity to shocks in fundamental (-0.01)
8
6
4
2
0
-2
-4
-6
-8
-10
-12
118
157
196
235
274
313
352
391
430
469
508
547
586
625
664
703
742
781
820
859
898
937
976
1
40
79
3.3 “Excess volatility”
The model discussed in the previous sections is driven by exogenous news in the
fundamentals and by the noise produced by the non-linear speculative dynamics embedded in
the model. As a result, the non-linear dynamics is capable of producing “excess volatility” in
the exchange rate, i.e. volatility that exceeds the volatility of the underlying fundamental. In
this section we analyse the sources of this excess volatility. We do this by computing the
noise to signal ratio in the simulated exchange rate. It gives a measure of how large the noise
produced by the non-linear dynamics is with respect to the exogenous volatility of the
fundamental exchange rate. We simulate this noise to signal ratio for different values of the
parameters of the model. We show the results in figures 5-7. We observe that the noise to
signal ratio is very sensitive to the extrapolation parameter of the chartists and to a lesser
degree to the transactions costs. We observe that the noise produced by the non-linear
dynamics can become very large relative to the volatility of the fundamental exchange rate
when the extrapolation parameter is large and when transactions costs are large. The noise to
signal ratio is less sensitive to the speed of adjustment in the goods market. Overall our model
is capable of generating volatility of the exchange rate that is much in excess of the volatility
of the underlying fundamental (Goodhart (1989) and Goodhart and Figlioli ( 1991)).
8Figure 5 Figure 6
ratio of noise to signal ratio of noise to signal
as a function of extrapolation by chartists (c=5; theta=-0.2) as a function of speed of adjustment (c=5; beta=2)
80 6
70 66.4
5
60
4
50
40 3
30
22.1 2
20
1
10
3.9
0.7
0 0
1 2 4 6 0.05 0.1 0.2 0.3 0.4
chartists' extrapolation speed of adjustment
Figure 7
noise to signal ratio as a function of transactions costs
(theta=-0.2; beta=2)
5.0
4.5
4.0
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
0 1 2 3 4 5
transactions costs
3.4 Fat tails
It is well known that the exchange rate changes do not follow a normal distribution. Instead it
has been observed that the distribution of exchange rate changes has more density around the
mean than the normal and exhibits fatter tails than the normal (see de Vries (2001)). Put
differently, the exchange rate returns typically have a kurtosis exceeding 3 and a measure of
fat tails (Hill index) ranging typically between 2 and 5. It implies that most of the time the
exchange rate movements are relatively small but that occasionally periods of turbulence
occur with relatively large exchange rate changes. However, it has been also detected that the
kurtosis is reduced under time aggregation. This phenomenon has been observed for most
exchange rates. We checked whether this is also the case with the simulated exchange rate
changes in our model.
The model was simulated using normally distributed random disturbances (with mean = 0 and
standard deviation = 1). We computed the kurtosis and the Hill index of the simulated
exchange rate returns. We computed the Hill index for 5 different samples of 2000
observations. In addition, we considered three different cut-off points of the tails (2.5%, 5%,
10%). We show the results of the kurtosis in table 3 and of the Hill index in table 4. We find
that for a broad range of parameter values the kurtosis exceeds 3 and the Hill index indicates
the presence of fat tails. This suggests that the non-linear dynamics of the model transforms
9normally distributed noise in the exchange rate into exchange rate movements with tails that
are significantly fatter than the normal distribution and with more density around the mean.
Thus our model mimics an important empirical regularity, i.e. that exchange rate movements
are characterised by tranquil periods (occurring most of the time) and turbulent periods
(occurring infrequently).
Table 3: Kurtosis index
Kurtosis of simulated exchange rate returns
theta -0.05 -0.1 -0.2 -0.3 -0.4
beta
0.5 2.99 3.21 8.48 8.58 11.84
1 2.98 3.09 2.14 15.17 9.66
1.5 2.92 2.96 3.78 9.27 7.03
2 2.85 2.88 3.31 4.84 6.05
2.5 2.11 2.64 2.8 3.86 5.68
3 1.44 1.85 2.49 3.54 5.64
Table 4 : Measure of fat tails : the Hill index
Median Hill index
(5 samples 2000 observations)
Parameter values Kurtosis 2.5% tail 5% tail 10%tail
c=5, beta=0.5, 8.48 2.42 2.66 2.93
theta=-0.2 (2.10 – 2.61) (2.50 – 3.15) (2.89 – 3.83)
c=5, beta=1, 8.1 2.14 2.35 2.65
theta=-0.2 (2.08 – 2.37) (2.21 – 2.45) (2.57 – 2.73)
c=5, beta=1, 15.17 1.54 1.62 2.14
theta=-0.3 (1.34 – 3.89) (1.46 – 2.00) (1.90 – 2.55)
c=5 , beta=1.5 9.27 7.36 1.82 1.70
theta=-0.3 (5.57 – 7.67) (1.58 – 3.13) (1.62 – 1.80)
4. Small and large shocks and the dynamics of the exchange rate
In linear models the size of the shocks does not affect the nature of the dynamics. In non-
linear models things are different. The size of the shocks matters. This is also the case in our
exchange rate model. In order to illustrate this, we simulated the model under two different
assumptions about the variance of the shocks in the fundamental exchange rate. In the first
case we assume low variance of these shocks, in the second case we assume a high variance
(ten times higher). The results of our simulations are presented in figures 8-9. (The
simulations shown here are representative for a wide range of parameter values).
10Figure 8: low variance of equilibrium exchange rate
market rate and fundamental sensitivity to parameter change theta
(c=3; theta=-0.2; beta=1.5 (c=3; theta=-0.2 and -0.21; beta=1.5)
9 9
8 8
7 7
6 6
5 5
4 4
3 3
2 2
1 1
0 0
1
1
49
97
145
193
241
289
337
385
433
481
529
577
625
673
721
769
817
865
913
961
46
91
136
181
226
271
316
361
406
451
496
541
586
631
676
721
766
811
856
901
946
Figure 9: high variance of equilibrium exchange rate
market rate and fundamental sensitivity to parameter change theta
(c=3; theta=-0.2; beta=1.5 (c=3; theta=-0.2 and -0.21; beta=1.5)
30 30
20 20
10 10
0 0
-10 -10
-20 -20
-30 -30
-40 -40
-50 -50
-60 -60
-70 -70
1
50
99
1
148
197
246
295
344
393
442
491
540
589
638
687
736
785
834
883
932
981
139
185
231
277
323
369
415
461
507
553
599
645
691
737
783
829
875
921
967
47
93
Two conclusions follow from a comparison of the low and high variance cases. First, in the
low variance case we observe sustained deviations from the equilibrium exchange rate; this is
not the case when the equilibrium exchange rate is subject to large shocks (compare left-hand
panels of figure 8 and 9). Second, the sensitivity to small changes in parameters is clearly
visible when the variance of the exchange rate is low (see right-hand panel of figure 8). When
this variance is high, no such sensitivity can be observed (right-hand panel of figure 9). It is
important to stress that the transactions cost band is the same in both cases. Thus, when the
shocks are small relative to the given band of transactions costs the movements of the
exchange rate show more complexity than when the shocks are large.
11This feature is also evident from a comparison of the noise to signal ratio for different
variances of the fundamental exchange rate. We show this in figure 10. We observe that when
the variance of the equilibrium exchange rate is low, a large part of the volatility of the
exchange rate is produced by the noise from the non-linear dynamics. For high variance the
noise is very small, implying that the exchange rate follows the fundamental rate very
closely 8 .
Figure 10
noise to signal ratio as a function of variance equilibrium rate
c=3; theta=-0.2; beta=1.5
1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10
0.00
0.5 1 2 4
variance equilibirum rate
The intuition of this result is that when the fundamental shocks are small the exchange rate
regularly switches from the dynamics inherent in the band to the one prevalent outside the
band. This non-linearity produces a lot of noise and complexity in the dynamics of the
exchange rate. When the shocks are large relative to transactions cost band the dynamics
outside the band mostly prevails, leading to a tighter link between the exchange rate and the
fundamental. This feature has also been found to hold empirically. In particular, it has been
found that the PPP-relationship holds much tighter in high inflation countries than in low
inflation countries (See De Grauwe and Grimaldi (2001)). Put differently, in high inflation
countries the link between the exchange rate and one of its most important fundamentals is
tighter than in low inflation countries. It may be argued that this is not yet evidence for the
model because the latter predicts that it is a high variance in the fundamental that produces
the tight link. However, there is also strong empirical evidence that the variance of inflation
increases with its level (see Okun (1971) Fischer (1982)). This strong correlation between the
level and the variance of inflation is illustrated in figure 11, which shows the average rate of
inflation and its standard deviation in a cross-section of 81 countries during 1971-1999.
8
It should be stressed that the total variability of the exchange rate in the high variance scenario is much larger than the total
variability of the exchange rate in the low variance scenario. The point is that in the high variance scenario almost all of the
variability of the exchange rate is explained by the (much higher) variability of the fundamental. This is not the case in the low
variance scenario where a large part of the variability of the exchange rate cannot be related to the variability of the underlying
fundamental.
12Figure 11
average yearly inflation and its volatility in cross-setion
of 81 countries (1971-99)
0.45
0.40
0.35
0.30
stdev inflation
0.25
0.20
0.15
0.10
0.05
0.00
-0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6
average inflation
5. Is chartism profitable?
In this section we analyse how profitable forecasting based on chartism is in relation with
fundamentalism. This analysis is important because particular forecasting rules will only
survive if they are profitable. If chartism turns out to be unprofitable, fewer and fewer agents
will use this technique, and it will disappear. In that case chartists should have no role in our
model.
In order to analyse this issue we simulated the model and asked the question how the profit
and loss accounts of chartists and fundamentalists evolve over time. We assumed that each of
them started with an initial capital of €1. When they expect the exchange rate to increase
(decrease) they buy (sell), and hold for one period. They repeat this operation each period.
We calculated the net present value of these profits and losses using a discount rate of 4%.
Results are shown in figure 12 where the present value of profits and losses are related to
different values of beta.
Figure 12
Present value of profits and losses fundis and chartists
c=5; theta=-0.2 (10000 periods)
70
fundis
60 chartists
50
present value
40
30
20
10
0
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
beta
13We observe the following. First, the cumulative profits of both chartists and fundamentalists
are positive.9 Second, for values of beta lower than 4 the chartists make profits higher than the
fundamentalists. However, for high values of beta the chartists’ rule looses its profitability
and the fundamentalists’ rule becomes much more profitable. This implies that we are
unlikely to observe chartists to use large extrapolation parameter values in their forecasting.
The next step was to analyse profits and losses under two different stochastic regimes. The
first one has a low variance of noise (same as in previous simulations). The second regime has
a variance ten times higher. Results are shown in figures 13 and 14. We see that chartism
becomes less profitable in a regime of high variance, while fundamentalism then becomes
more profitable. It is worthwhile to note that this result is consistent with the results obtained
in the previous section, where we showed that in a high variance regime the link between
fundamentals and the market exchange rate is tighter than in the low variance regime. Thus, it
is not surprising that in a high variance regime the fundamentalists’ forecasting rule is
relatively profitable. This result also implies that we should observe more chartists in the low
variance currency markets than in the high variance markets. We leave it for further research
to verify the empirical validity of this hypothesis 10 .
Figure 13 Figure 14
Present value profits and losses chartists Present value profits and losses fundis
c=5; theta=-0.2 (after 10000 periods) c=5; theta=-0.2 (after 10000 periods)
20 80
70
15
60
10
present value
50
5 40
30
0
20
-5 low var low var
10
high var high var
-10 0
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
beta beta
6. Conclusion
In this paper we analysed the workings of a simple non-linear exchange rate model in which
agents hold different beliefs about the underlying model. We distinguished between chartists
and fundamentalists, where the chartists apply a positive feedback rule and the
fundamentalists a negative feedback rule. The non-linearities in the model originate from
transactions costs and from the existence of non-linear adjustment dynamics in the goods
market.
Our main results can be summarised as follows. First, the simple non-linear structure of the
model is capable of generating a very complex exchange rate dynamics. This implies that
9
We assume implicitly that the total gains (losses) of these forecasters must have a counterpart loss (gain) borne by other agents
(e.g. exporters and importers) outside the model. We assume that the gains and losses of the forecasters are small in a
macroeconomic sense so that spread over many agents in the rest of the economy they do not affect the latters’ behaviour.
10
Recently Cheung, et al (2002) found such evidence.
14small shocks in the equilibrium exchange rate lead to very different time-paths of the
exchange rate.
Second, our model is capable of explaining most empirical puzzles associated with exchange
rate movements. The first puzzle is that the market exchange rate can deviate substantially
and for relatively long periods of time from its fundamental value (“disconnect puzzle”). We
showed that such disconnections are a natural outcome of the non-linear dynamics in our
simple model. There is no need to invoke exogenous events and special factors to explain why
exchange rates deviate from their fundamental values. It should also be noted that our model
generates these disconnections even in the absence of deterministic chaos. In other words we
do not need to invoke chaos to explain disconnections.
Another empirical puzzle observed in exchange rate economics is the frequent occurrence of
“regime shifts”, i.e. structural breaks in the relation between the exchange rate and the
fundamentals. This phenomenon has first been noted in the celebrated studies of Meese and
Rogoff (1982) 11 . It is now customary to explain these structural breaks by changes in the
policy regime. Our model provides an alternative explanation. The non-linear dynamics
embedded in the model produces endogenous regime shifts that change the link between the
exchange rate and its fundamentals. These structural breaks can be triggered by very small
changes in parameters, or by small errors in the estimates of these parameters by agents who
forecast the future exchange rate. Thus, in a non-linear world, structural breaks in the link
between the exchange rate and its fundamentals occur naturally even when no changes occur
in the policy regime.
Third, we found that our simple non-linear dynamic model can generate “excess volatility” in
the exchange rate. The size of this excess volatility crucially depends on the degree of
extrapolation applied by chartists and on the size of the transactions cost band.
Fourth, the model is also capable of generating another empirical regularity, i.e. the existence
of fat tails in the distribution of the exchange rate returns. Thus the model mimicks an
empirical phenomenon that has been widely observed in exchange rate economics, i.e. that
tranquil periods (which occur most of the time) alternate with turbulent periods (which occur
infrequently).
Fifth, we found that the size of shocks to the underlying fundamental exchange rate matters
for the dynamics of the exchange rate. More specifically, we found that when these shocks are
small relative to the size of the transactions cost band, the phenomena just described will tend
to be prevalent. That is, in regimes of low shocks relative to the transactions cost band, the
exchange rate movements are complex, and can even be chaotic. In such a regime exchange
rates deviate substantially from the underlying fundamentals and frequent structural breaks in
the link between the fundamentals and the exchange rate are observed. The latter occur in the
absence of changes in the policy regime.
Finally, we checked whether the forecasting rules used by chartists and fundamentalists are
profitable. We found that for a broad range of parameter values both rules are profitable.
Some implications of these findings are the following. The exchange rates of the major
currencies are subject to relatively small shocks in the underlying fundamentals (e.g. inflation
differentials are almost zero). Compared to these shocks the transactions costs can be said to
be relatively large (see Obstfeld & Rogoff (2000) on this), i.e. a large part of goods and
services are non-traded (or difficult to trade) because the cost of shipping them across borders
is quite high. Thus the regime confronted by the exchange rates of the major industrialised
countries comes close to the regime we have identified to be the one producing complexity,
11
For more recent evidence see De Grauwe and Vansteenkiste (2001).
15speculative noise, and structural breaks between exchange rates and underlying fundamentals.
Put differently, the movements of the exchange rates of the industrialised countries are likely
to be clouded by a non-linear speculative dynamics that makes it difficult if not impossible to
explain this or that movement of these exchange rates. In contrast the exchange rates of high
inflation countries experience large shocks in the fundamentals. As a result, the movements of
the exchange rates of these countries can be explained much better by movements in
underlying fundamentals (e.g. inflation differentials).
The results of our paper make it easier to understand why it will remain difficult, if not
impossible to find (fundamental) logic in the movements of the exchange rates of major
currencies that are subject to relatively low nominal disturbances. However, our inability to
understand why, say, the dollar moved up against the euro during the 1999-2000 does not
prevent analysts from developing exotic theories explaining these movements. Probably this
has to do with the fact that the human mind abhors the emptiness created by its inability to
understand. It is no surprise therefore that new explanations based on fundamentals are
created, and will continue to be created for each and every new turn of the exchange rate.
16Appendix : The theoretical model
In this appendix we present the exchange rate model that was used to perform the simulations
reported in this paper. We start with a well-known model of the exchange rate, which is often
used in the literature. We then introduce heterogeneous agents who use this model as a
benchmark to define their beliefs about the future exchange rate.
We start from the determination of the exchange rate as follows:
st = f t + α[Et s t +1 − s t ] (1)
where f t represents the fundamentals in period t, st is the exchange rate in period t, st+1 is the
exchange rate in period t+1, E is the expectations operator. Underlying the fundamental one
could specify a whole model of the economy, e.g. a monetary model, or a more elaborate one
like the Obstfeld-Rogoff new open economy macro model (Obstfeld and Rogoff(1996)). We
leave this for further research. Here we concentrate on the simplest possible exchange rate
modelling. For the sake of simplicity, we assume that the fundamentals are determined
exogenously 12 .
Equation (1) can be rewritten as follows:
α
st =
1
ft + [Et s t+1 ] (2)
1+α 1+α
We use this model to define the fundamental equilibrium exchange rate. This is the rational
expectations solution of equation (2). It will be used as a benchmark against which the beliefs
of different agents are measured.
In the absence of bubbles the fundamental solution to (2) is given by
i
1 ∞ a
st* = ∑ Et f t + i
1 + a i = 0 1 + a
(3)
For the sake of simplicity we will assume that ft follows a random walk process without drift.
We then find the following fundamental solution of the exchange rate :
st* = f t (4)
We now introduce the assumption that the agents have heterogeneous beliefs and we classify
them according to their beliefs. Let us assume that there are Nh individuals of type h belief
(where ΣNh = N). We can then characterize the beliefs of type h agents as follows13 :
Eh , t s t + 1 = st* + g h ( s t − 1 , s t − 2 ,...) (5)
where Eh,t represents the expectations operator of type h agent at time t. Thus agents’ beliefs
can be classified depending on how they view the process by which the market price will
grope towards the fundamental exchange rate s * t. They all use information on past exchange
rates to forecast these future developments.
12
This means that we assume implicitly that there is no feedback from the exchange rate to the goods market. We leave it for
further research to explore this possibility.
13
See Brock and Hommes (1998) for such a formulation
17The market expectation can then be written as follows:
H H
Et st +1 = ∑ n h E h, t [s t +1 ] = st* + ∑ n h g h (st −1 , st − 2 ,...) (6)
h =1 h =1
Note that n h = Nh /N, so that n h can be interpreted as the weight of agents of type h in the
market.
The realised market rate in period t+1 equals the market forecast made at time t plus some
white noise error (i.e. the news that could not be predicted at time t) :
H
st +1 = s *t + ∑ nh g h (s t −1 , s t − 2 ,...) + εt +1 (7)
h =1
In the previous discussion the nature of the beliefs of agents was specified in very general
terms. We further simplify the model by assuming that there are only two types of agents in
the foreign exchange market, which we will call fundamentalists and chartists 14 .
The fundamentalists base their forecasts on a rule like in equation (5), i.e. they compare the
past market exchange rates with the fundamental rate and they forecast the future market rate
to move towards the fundamental rate. In this sense they follow a negative feedback rule 15 .
We will make the additional assumption that they expect the speed with which the market rate
returns to the fundamental rate to be determined by the speed of adjustment in the goods
market.
As pointed out earlier, there is an increasing amount of empirical evidence indicating that the
speed of adjustment in the goods market follows a non-linear dynamics, i.e. the speed with
which prices adjust towards equilibrium depends positively on the size of the deviation from
equilibrium. We will assume that this adjustment process is quadratic in nature16 .
Fundamentalists take this non-linear dynamic adjustment into account in making their
forecast. This leads us to specify the following rule for the fundamentalists:
(
E f ,t (∆st +1 ) = −ψ st −1 − st*−1 ) (8)
where Ef,t is the forecast made in period t by fundamentalists and ψ is a function of the size of
the deviation from the fundamental variable. We assume the following simple specification
ψ = θ st −1 − s *t −1
where θ > 0
Thus when the size of the deviation from equilibrium is large the fundamentalists expect a
faster speed of adjustment towards the fundamental rate than when the size of the deviation is
small. The economics behind this non-linear specification is that in order to profit from
14
This way of modelling the foreign exchange market was first proposed by Frankel and Froot (1988). It was further extended by
De Long et al. (1990) and De Grauwe et al. (1993) and more recently Kilian and Taylor (2001). For evidence about the use of
chartism see Taylor and Allen (1992).
15
Note that this is also the approach taken in the Dornbusch model.
16
See Kilian and Taylor (2001). See also De Grauwe and Grimaldi (2001) in which we showed that a quadratic specification fits
the data rather well.
18arbitrage opportunities in the goods market, some fixed investment must be made, e.g. trucks
must be bought, planes be chartered, etc. These investments become profitable with
sufficiently large deviations from the fundamental exchange rate. Note that we do not model
the goods market explicitly but we assume that in order to form their expectations about the
exchange rate, the fundamentalists take into account the dynamics of the goods market and
the speed of adjustment of goods prices.
The chartists are assumed to follow a positive feedback rule, i.e. they extrapolate past
movements of the exchange rate into the future. Their forecast is written as:
T
Ec, t (∆st +1 ) = β ∑ αi ∆s t − i (9)
i =1
where Ec,t is the forecast made by chartists using information up to time t; ∆st is the change in
exchange rate.
As can be seen, the chartists compute a moving average of the past exchange rate changes and
they extrapolate this into the future exchange rate change. The degree of extrapolation is
given by the parameter β. Note that in contrast to the general rule as given by equation (5)
(and also in contrast to fundamentalists) they do not take into account information concerning
the fundamental exchange rate. In this sense they can be considered to be pure noise traders17 .
It is important to check if the model is internally consistent. In particular, the chartists’
forecasting rule must be shown to be profitable within the confines of the model. If these rules
turn out to be unprofitable, they will not continue to be used. They can then also not been
maintained in the model. We will therefore check whether these rules are indeed profitable.
In a similar logic as in equation (7) the market exchange rate can now be written as
T
∆st +1 = −n ftθ st −1 − st*−1 ( st −1 − st*−1 ) + nct β ∑ ai ∆s t − i + εt +1
i =1
In the simulations of the model we assume that the weights n ft and n ct are constant. We set
them equal to 0.5, not because we think this is realistic but to see how far the simplest
possible model goes in explaining the exchange rate dynamics. In our future research we will
allow the weights given to fundamentalists and chartists to react endogenously to the
profitability of these forecasting rules.
As stressed in the empirical and theoretical literature, transaction costs are important to
explain the dynamics of the adjustment. Therefore, we will develop a version of the previous
model represented by equations (1)-(9) in which the transaction costs play a role.
We take the view that if transaction costs exist, the fundamentalists will take this information
into account. Therefore, if the exchange rate is within the transaction costs band the
fundamentalists will behave differently than if the exchange rate moves outside the
transaction costs band.
Consider the first case, when the exchange rate deviations from the equilibrium value are
smaller than the transaction costs, then the fundamentalists know that arbitrage in the goods
market does not apply. As a result, they expect the changes in the exchange rate to follow a
white noise process εt. The best they can do is to forecast no change. More formally,
17
See De Long et al. (1990)
19when st −1 − s *t −1 < C , then EFt (∆st+1) = 0.
In the second case, when the exchange rate deviation from its fundamental value is larger than
the transaction costs C (assumed to be of the ‘iceberg’ type). Then the fundamentalists follow
the same forecasting rule as in equation (8). More formally,
when st −1 − st*−1 > C holds, then equation (8) applies.
This formulation implies that when the exchange rate moves outside the transaction costs
band, market inefficiencies other than transaction costs continue to play a role. As a result,
these inefficiencies prevent the exchange rate from adjusting instantaneously. In our model
these inefficiencies are captured by the fact that the speed of adjustment in the goods market
is not infinite (equation (8)).
20References
Baxter, M., Stockman, 1989, A., Business Cycles and the Exchange Rate Regime. Some
International Evidence, Journal of Monetary Economics, 23, may 377-400.
Cheung Y. and Lai K., 2000. “ On the purchasing power parity puzzle”. Journal of
International Economics, 52 .
Copeland, L., 2000, Exchange Rates and International Finance, 3rd ed., Prentice Hall.
De Grauwe, P. , Dewachter, H., and Embrechts, 1993, M., Exchange Rate Theories.
Chaotic Models of the Foreign Exchange Markets, Blackwell.
De Grauwe, P., and Grimaldi, M., 2001, Exchange Rates, Prices and Money: A Long Run
Perspective, International Journal of Finance and Economics, 6, no. 4, pp. 289-314.
De Grauwe, P., and Vansteenkiste, I., 2001, Exchange Rates and Fundamentals. A Non-
linear Relationship?, CESifo Working Paper, no. 577, October.
de Vries, C., 2000, Fat tails and the history of the guilder, Tinbergen Magazine, 4, Fall, pp.
3-6.
De Long, J., Bradford, B., Schleiffer and Summers, L., 1990, Noise Trader Risk in
Financial Markets, Journal of Political Economy.
Dornbusch R., 1976, “Expectations and exchange rate dynamics”, Journal of Political
Economy 84.
Dumas B., 1992, Dynamic equilibrium and the real exchange rate in a spatially separated
world, Review of financial studies 5 (2) , 153-180.
Engel C., 2000, “ Long run PPP may not hold after all”, Journal of International
Economics, 57.
Engel C. and Morley J., 2001, The adjustment of prices and the adjustment of the
exchange rate, Discussion paper, Department of Economics, University of Wisconsin.
Evans, M., and Lyons, R., 1999, Order Flow and Exchange Rate Dynamics, NBER
Working Paper, no. 7317.
Fischer S., 1982, Relative prices and inflation in the United States and Germany, European
Economic Review, No. 18.
Flood, R, and Rose, A., 1995, Fixing the Exchange Rate Regime: A virtual Quest for
Fundamentals, Journal of Monetary Economics, 36, August, 3-37.
Frankel, J., and Froot, K., 1986, The Dollar as a Speculative Bubble: A Tale of
Fundamentalists and Chartists, NBER Working Paper, no. 1963.
Frenkel J., 1981, The collapse of purchasing power parity during the 1970s, European
Economic Review 7.
Goodhart, C., 1989, News and the Foreign Exchange Market, LSE Financial Markets
Group Discussion paper, 71.
Goodhart, C., and Figliuoli, L., 1991, Every Minute Counts in the Foreign Exchange
Markets, Journal of International Money and Finance, 10, 23-52.
Hallwood, P., MacDonald, R., 1994, International Money and Finance, 2nd ed., Blackwell,
Oxford.
Isard, P., 1995, Exchange Rate Economics, Cambridge University Press.
Kilian L. and M. Taylor, 2001, Why is it So Difficult to Beat the Random Walk Forecast
of Exchange Rates? Mimeo, University of Warwick, pp. 29.
21Kurz, M., 1994, On the Structure and Diversity of Rational Beliefs, Economic Theory, 4,
877-900.
Kurz, M., and Motolese, M., 2000, Endogenous Uncertainty and Market Volatility, mimeo,
Stanford University.
LI K., 1999, Testing symmetry and proportionality in PPP : A panel data approach.
Journal of Business and Economic Statistics 17 (4) , 409-418.
Lux T., 1998, The socio-economic dynamics of speculative markets: interacting agents,
chaos, and fat tails of return distributions, Journal of Economic Behaviour and
Organisation, vol.33.
Lux T., Marchesi M., 2000, Volatility clustering in financial markets: a microsimulation
of interacting agents, International Journal of Theoretical and Applied Finance.
Lyons, R., 2001, The Microstructure Approach to Exchange Rates, MIT Press, Cambridge,
Mass.
Meese, R., and Rogoff, 1983, Empirical Exchange Rate Models of the Seventies: Do they
Fit Out of Sample ?, Journal of International Economics, 14, 3-24.
Michael P., Nobay R., and Peel A., 1997, Transaction costs and non-linear adjustment in
real exchange rates: an empirical investigation, Journal of Political Economy 105 (4), 862-
879.
Obstfeld, M. and Rogoff, K., 1996, Foundations of International Macroeconomics, MIT
Press, Cambridge, Mass.
Obstfeld, M., and Rogoff, K., 2000, The Six Major Puzzles in International
Macroeconomics: Is there a Common Cause?, NBER Working Paper no. 7777, July.
Okun A., 1971, The mirage of steady inflation, Brooking papers on Economic activity .
Rogoff, K., 1996, The purchasing power parity puzzle, Journal of Economic Literature,
34, June, 647-668.
Taylor, M., and Allen, H., 1992, The Use of Technical Analysis in the Foreign Exchange
Market, Journal of International Money and Finance, 11, 304-14.
Taylor M., Peel D., and Sarno L., 2001, Non-linear mean reversion in real exchange rates:
towards a solution to the purchasing power parity puzzles. CEPR discussion paper no
2658.
22You can also read
